Voice Engines: System Voices, Cloud AI Voices, and Bring Your Own API Key

The "voices" in Soniva come from three sources, with completely different costs and prerequisites. Understanding all three is how you know when credits are deducted and when it's completely free.

---

1. System Built-in Voice (Free forever, no credits)

This calls the reading engine already built into your own device/browser:

• Completely free, no limit on count or duration, consumes no credits.
• The voices are decided by your system; Soniva can't add to or change the selectable voices.
• Quality is usually below cloud AI voices, but it's sufficient for the "read this text aloud" need.
• No subscription and no login required. When logged out, sessions exist only in the current browser; once logged in, sessions sync to the cloud and are visible across devices (sync doesn't require a subscription).

---

2. Cloud AI Voices (require subscription, billed in credits)

Soniva integrates 8 cloud voice brands uniformly, paying upstream through Soniva's own accounts, so using them requires an active subscription + spending credits:

Each brand also has multiple model tiers with big price differences — the cheapest and most expensive tiers differ by roughly 42×. Full pricing is in the credit rate table.

No need to memorize the price list when choosing a voice: after picking a brand and model in "TTS Settings," the estimated consumption rate appears right below the dropdown (e.g. "~16 credits / 1,000 chars").

Voice Lists Are Pulled Live, Not Hard-Coded

For Doubao Speech, MiniMax, ElevenLabs, Microsoft Azure, and Cartesia, the voice lists are pulled live from the upstream interface by Soniva every time — when upstream adds voices, you'll see them in Soniva directly, no need to wait for a Soniva release. So how many voices there are, go by what you actually see in the product; this document doesn't list a fixed number (it would go stale fast).

Alibaba Cloud, Tencent Cloud, and Baidu don't provide a public voice-list interface, so they use Soniva's built-in static lists.

Free Previews

Even without a subscription — or even without logging in — you can preview cloud voices for free:

• Alibaba Cloud / Doubao / MiniMax / Tencent Cloud / Baidu / Microsoft: each provides 1 representative voice + 1 fixed script, with a 5-per-brand-per-day limit (counted by IP, counted separately across the 6 brands), and the script can't be changed.
• ElevenLabs / Cartesia: these two brands' voice lists already come with official pre-made preview audio that plays directly without going through Soniva's backend, so there's far more of it and no count limit.

How to find the freely playable voice: brands with a small dot marker on the brand selector have a free preview voice; when you switch to that brand, Soniva automatically switches the model selector to the tier containing the free voice (e.g. Doubao's free voice is under 1.0, while the default model is 2.0).

We know "a brand with hundreds of voices where only one is freely playable" isn't a great experience; the long-term goal is to make all voices directly playable like ElevenLabs does. That's in the plan, but not built yet.

Per-Reading Length Limit

• Reading with Soniva credits: per-request cap of 500 characters.
• Reading with your own API key (BYOK): per-request cap of 4,000 characters.

For text above the cap, Soniva auto-splits it into multiple segments and reads them in sequence — you don't need to split it yourself, you'll just see multiple records in the billing ledger. Also, Doubao's small model (BV voices) splits into finer ~280-character segments because upstream bills per call, which directly affects its actual charge — see Detail 3 of the rate table.

---

3. Bring Your Own API Key (BYOK, no credits)

If you already have a developer account on platforms like Alibaba Cloud, Volcano Engine, or ElevenLabs, you can enter your own API key in Soniva and requests go directly through your own account:

1. Open the personal menu in the bottom-left, select Settings
2. Switch to the Voice API tab
3. Enter the corresponding brand's API key and save

Soniva deducts no credits with this method; the provider bills your account directly. It suits users who've already bought bundles, or whose volume makes opening their own accounts more economical.

API Key Security

• An entered key is saved to your current browser's local storage; if you check "Sync to account," it's also stored under your account encrypted with AES-256-GCM — the database holds ciphertext, the decryption key exists only server-side, and neither the browser nor the database can get the plaintext.
• Synthesis requests are forwarded by Soniva's server to the upstream brand on your behalf (to handle each vendor's differing signature/auth protocols); your key is used only for your own requests, never for any other user.
• Logging out clears all locally saved keys. You can also manually clear them on the same settings page anytime.

---

How to Choose

---

Related Documents

• Ways to read aloud: offline voice packs, local CosyVoice and system voices (desktop)
• Eight Brands at a Glance: Positioning and Selection Advice
• TTS Workbench Guide
• Long-Text Reading: How Auto-Segmentation Works
• Full Credit Rate Table
• How to Choose a Subscription Tier
• What a Given Credit Budget Buys